| Abstract [eng] |
The main purpose of the research is to help investors to make the most accurate decisions about buying, selling or holding shares. In this study it is predicted that the stock prices of the three companies will change one day ahead, using historical data on closing prices (stock trading data and technical indicators), and the three Baltic stock market indices will change one month ahead when various macroeconomic indicators are chosen. Predicting is implemented by adapting machine learning methods and using classification when the values of the target columns are as follows: \"price / index will rise\" and \"price / index will not rise\". This is preceded by the assumption that prediction of volatility can be more accurate if significant variables are selected and this is done using different sensitivity analysis methods. When predicting stock price volatility of companies, in the cases of Lithuania and Estonia the selection of significant variables did not influence, and in the case of Latvia the partial correlation method was useful, as then the accuracy of classification was improved. AUC estimates (0.836, 0.824, and 0.805) of the best models showed that only historical stock price data can provide good predicting results one day ahead when using the Support Vector Machine algorithm. Predicting the volatility of stock market indices one month ahead, the AUC values of models were not high for any country. However, the selection of significant variables by applying sensitivity analysis methods had an impact on the accuracy of classification by at least a few percent. The data drift in one or two periods (months) ahead (assuming that macroeconomic indicators may not immediately affect index changes) also often had positive impact on the accuracy of classification. In this study, the best classifiers are: Random Forests, Support Vector Machine and Decision Tree. The most significant macroeconomic variables selected by country were: wages, employment of population, general government debt (Lithuania); wages, construction cost element price index (Latvia); industrial output price index, one and three month EURIBOR interest rates, gross national income (Estonia). |
